CN108961397B - A Simplified Method for Tree-Oriented 3D Point Cloud Data - Google Patents

Abstract

The invention provides a tree-oriented three-dimensional point cloud data simplifying method, which constructs a three-dimensional cubic grid, calculates homogeneous connection weights of the network and neighbor grids, and sequentially eliminates some redundant points according to the size of the homogeneous connection weights so as to simplify the three-dimensional point cloud data of trees. The method simplifies the data and maintains the vegetation structure expressed by the three-dimensional point cloud data of the original tree as much as possible, and has higher practical application value in the field of tree three-dimensional modeling.

Description

A Simplified Method for Tree-Oriented 3D Point Cloud Data

technical field

The invention belongs to the technical field of spatial positioning, remote sensing and communication network fusion, and in particular relates to a simplified method for tree-oriented three-dimensional point cloud data, which utilizes Internet, big data and cloud computing technologies to process three-dimensional data of plants.

Background technique

At present, laser scanners, lidars and other equipment can quickly scan objects in 3D, and these devices obtain point cloud data composed of a large number of 3D space points. Using these point cloud data can quickly and cost-effectively carry out 3D modeling of objects, which has important practical application significance.

However, dense spatial scanning brings relatively fine detail information of objects, but also brings great computational and storage burdens. Simplify, keep the features of the original objects and try to reduce the number of 3D space points in the point cloud dataset. The simplification of 3D point cloud data is very important to obtain 3D objects with less data and easier to calculate and display.

The current methods and technologies are mostly focused on: finding the rules of planes and surfaces in objects through clustering algorithms, octree segmentation and search, and deleting or merging redundant and repeated points. These techniques are more effective in simplifying relatively regular man-made buildings, man-made objects, and objects with large-area planar structures. However, when dealing with trees, due to the leaves and branches contained in trees, it is difficult to find surface and surface patterns on these objects, and some relatively important branches and overall morphology are discarded by forced clustering and octree segmentation. information, so current algorithms are less efficient for simplifying tree 3D point cloud data.

The task of the present invention is to propose a simplification method for 3D point cloud data of coniferous trees, which can simplify the number of 3D space points according to the characteristics of the tree point cloud.

Invention content:

Aiming at the problems existing in the prior art, the present invention provides a simplified method for tree-oriented 3D point cloud data. The method constructs a 3D cubic grid and calculates the homogeneous connection weight between the network and the neighbor grid. The size of the weights is used to sequentially remove some redundant points and simplify the 3D point cloud data of trees.

A simplified method for tree-oriented three-dimensional point cloud data according to the present invention, comprising the following steps:

S1. Input the tree point cloud dataset PointDataSet containing n three-dimensional space points, input the grid side length width, input the number of homogenized clusters clusternum, and input the reduction percentage percent:

S101. Input a tree point cloud data set PointDataSet, PointDataSet is a data set composed of three-dimensional space points {p ₁ , p ₂ ,..p _n } The data set contains n three-dimensional space points; any one of the three-dimensional points pi={ px,py,pz},px is the coordinate of the point on the x-axis, py is the coordinate of the point on the y-axis, pz is the coordinate of the point on the z-axis;

S102, count all points in the PointDataSet, and obtain the maximum value of the x coordinate xmax, the minimum value of the x coordinate xmin, the maximum value of the y coordinate ymax, the minimum value of the y coordinate, the maximum value zmax of the z coordinate, and the minimum value zmin of the z coordinate of all points in the PointDataSet. ;

S103, the x-axis span xdistance=xmax-xmin of all points in the PointDataSet, the y-axis span ydistance=ymax-ymin, the z-axis span zdistance=zmax-zmin;

S104, input the grid side length width, and input the number of homogenized bunches clusternum;

The default value of grid side length width is xdistance/1000;

The default value of clusternum, the number of homogenized clusters, is 100;

S2. Cut the three-dimensional space occupied by the PointDataSet into multiple cubic grids according to the side length of width to form a cubic grid list CubeList:

S201, cube grid list CubeList=empty collection;

S202, x iterator xcounter=xmin;

S203, y iterator ycounter=ymin;

S204, z iterator zcounter=zmin;

S205, a cubic grid Cube is established, and the Cube includes the following properties:

The spatial scope corresponding to Cube is from (xcounter, ycounter, zcounter) to (xcounter+width, ycounter+width, zcounter+width)

The coordinates of the center point of the Cube:

cp=(xcounter+width/2,ycounter+width/2,ycounter+width/2)

Cube's point density density=0; Cube's membership category=0; Cube's homogeneous weight priority=0;

S206, the point density of Cube=the number of all spatial points contained in the Cube space in the PointDataSet;

S207, add Cube to CubeList;

S208, zcounter=zcounter+width;

S209, if zcounter>=zmax, go to S210, otherwise go to S205;

S210, ycounter=ycounter+width;

S211, if ycounter>=ymax, go to 212, otherwise go to S204;

S212, xcounter=xcounter-1;

S213, if xcounter>=xmax, go to S214, otherwise go to S203;

S214, the calculation process of this step ends;

S3. Calculate the category of each Cube in the CubeList, and obtain the distance index distanceindex:

S301, take the point density density of each Cube in the CubeList as the input data, divide the input data into clusternum categories through the k-means algorithm, and obtain clusternum cluster centers c[1], c[2],...c[ clustenum]; where c[1] represents the cluster center numbered 1, c[2] represents the cluster center numbered 2, c[clustenum] represents the cluster center numbered clusternum; c[i] represents the i-th cluster center a cluster center;

S302, for each Cube in the CubeList, calculate the distance between the density of the Cube and the clusternum cluster centers, find the cluster center with the closest distance from the density, and store the number of the cluster center in the category of the Cube's generic relationship. middle;

S303, create an array CloseDistance=empty array;

S304, for each three-dimensional space point in the PointDataSet, find another three-dimensional space point closest to the point and add the distance value to the CloseDistance;

S305, calculate the mean avgdistacne of CloseDistance;

S307, count the mean avgcatalog of the categories of all Cubes in the CubeList;

S308, obtain the distance index distanceindex, and its calculation formula is as follows:

S4. Calculate the priority of the homogeneous weights of each Cube in the CubeList:

S401, the following steps are performed for each Cube in the CubeList;

S402, grid currentCube to be processed at present=take out a Cube from the CubeList;

S403, the current center point centerpoint=coordinate cp of the center point position of the currentCube to be processed;

S404, neighbor list neighborlist=calculate the distance from the cp of all Cubes in the CubeList to the centerpoint, and extract all Cubes whose distance is less than or equal to distanceindex;

S405, allnum=number of cubes in neighborlist;

S406, the homogenization counter scounter=0, the homogenization counter ycounter=0, the iteration counter counter=1;

S407, testCube=take out the counter-th Cube in neighborlist;

S408, diff=ABS (the affiliation category of the currentCube—the affiliation category of the testCube);

Among them, ABS represents the absolute value of calculation;

S409, if the diff is greater than 2, then ycounter=ycounter+1, otherwise counter=scounter+1;

S410, counter=counter+1;

S411, if counter is greater than allnum, go to S412, otherwise go to S405;

S412, calculate the homogeneous connection permission pri, and the corresponding formula is as follows:

S413. The homogeneous weight priority=pri of the currentCube;

S5, continuously remove the three-dimensional space points from the PointDataSet according to the homogeneous connection weight until the reduction percentage is reached:

S501, std=calculate the standard deviation of the priority of the homogeneous weights of all cubes in the CubeList;

S502, the rotation value circle=std/clusternum;

S503, sort all cubes in the CubeList from small to large according to the priority value;

S504, the smallest grid to be processed FirstCube=takes out the Cube ranked first in the CubeList;

S505, ptlist=in the PointDataSet, take out all the three-dimensional space points contained in the FirstCube space;

S506, if the number of points in the ptlist is less than 8, go to S507, otherwise go to S508;

S507, delete FirstCube from the CubeList, and go to S510;

S508, find the point closest to the coordinate cp of the center point of the FirstCube in the ptlist, and delete the point from the PointDataSet;

S509, the homogeneous connection weight priority of FirstCube=the homogeneous connection weight priority+circle of FirstCube;

S510, currentN=calculate the number of three-dimensional space points currently included in the PointDataSet;

S511, if (n-currentN)/n is less than percent, go to S503, otherwise go to S512;

S512, output the PointDataSet as the simplified result.

The beneficial effects of the present invention are as follows: a simplified method for tree-oriented three-dimensional point cloud data is provided. size to remove some redundant points in sequence and simplify the 3D point cloud data of trees. The present invention simplifies data and keeps the vegetation structure expressed by the original three-dimensional point cloud data of trees as much as possible, and has high practical application value in the field of three-dimensional modeling of trees.

Description of drawings

Fig. 1 is the dendrogram that embodiment 2 consists of three-dimensional space points;

2 is a simplified three-dimensional pine tree diagram obtained by the method of the present invention in Embodiment 2;

Fig. 3 is the dendrogram that embodiment 3 is formed by three-dimensional space point;

FIG. 4 is a simplified three-dimensional pine tree diagram obtained by the method of the present invention in Example 4. FIG.

Detailed ways

The specific embodiments of the present invention are described by the following examples, but those skilled in the art should understand that this is only an illustration, the protection scope of the present invention is defined by the appended claims, and those skilled in the art should not On the premise of departing from the principle and essence of the present invention, various changes or modifications can be made to these embodiments, and these changes and modifications all fall within the protection scope of the present invention.

Example 1

1. A simplified method for tree-oriented 3D point cloud data, comprising the following steps:

S1. Input the tree point cloud dataset PointDataSet containing n three-dimensional space points, input the grid side length width, input the number of homogenized clusters clusternum, and input the reduction percentage percent:

S101. Input a tree point cloud data set PointDataSet, PointDataSet is a data set composed of three-dimensional space points {p ₁ , p ₂ ,..p _n } The data set contains n three-dimensional space points; any one of the three-dimensional points pi={ px,py,pz},px is the coordinate of the point on the x-axis, py is the coordinate of the point on the y-axis, pz is the coordinate of the point on the z-axis;

S102, count all points in the PointDataSet, and obtain the maximum value of the x coordinate xmax, the minimum value of the x coordinate xmin, the maximum value of the y coordinate ymax, the minimum value of the y coordinate, the maximum value zmax of the z coordinate, and the minimum value zmin of the z coordinate of all points in the PointDataSet. ;

S103, the x-axis span xdistance=xmax-xmin of all points in the PointDataSet, the y-axis span ydistance=ymax-ymin, the z-axis span zdistance=zmax-zmin;

S104, input the grid side length width, and input the number of homogenized bunches clusternum;

The default value of grid side length width is xdistance/1000;

The default value of clusternum, the number of homogenized clusters, is 100;

S2. Cut the three-dimensional space occupied by the PointDataSet into multiple cubic grids according to the side length of width to form a cubic grid list CubeList:

S201, cube grid list CubeList=empty collection;

S202, x iterator xcounter=xmin;

S203, y iterator ycounter=ymin;

S204, z iterator zcounter=zmin;

S205, a cubic grid Cube is established, and the Cube includes the following properties:

The spatial scope corresponding to Cube is from (xcounter, ycounter, zcounter) to (xcounter+width, ycounter+width, zcounter+width)

The coordinates of the center point of the Cube:

cp=(xcounter+width/2,ycounter+width/2,ycounter+width/2)

Cube's point density density=0; Cube's membership category=0; Cube's homogeneous weight priority=0;

S206, the point density of Cube=the number of all spatial points contained in the Cube space in the PointDataSet;

S207, add Cube to CubeList;

S208, zcounter=zcounter+width;

S209, if zcounter>=zmax, go to S210, otherwise go to S205;

S210, ycounter=ycounter+width;

S211, if ycounter>=ymax, go to 212, otherwise go to S204;

S212, xcounter=xcounter-1;

S213, if xcounter>=xmax, go to S214, otherwise go to S203;

S214, the calculation process of this step ends;

S3. Calculate the category of each Cube in the CubeList, and obtain the distance index distanceindex:

S301, take the point density density of each Cube in the CubeList as the input data, divide the input data into clusternum categories through the k-means algorithm, and obtain clusternum cluster centers c[1], c[2],...c[ clustenum]; where c[1] represents the cluster center numbered 1, c[2] represents the cluster center numbered 2, c[clustenum] represents the cluster center numbered clusternum; c[i] represents the i-th cluster center a cluster center;

S302, for each Cube in the CubeList, calculate the distance between the density of the Cube and the clusternum cluster centers, find the cluster center with the closest distance from the density, and store the number of the cluster center in the category of the Cube's generic relationship. middle;

S303, create an array CloseDistance=empty array;

S304, for each three-dimensional space point in the PointDataSet, find another three-dimensional space point closest to the point and add the distance value to the CloseDistance;

S305, calculate the mean avgdistacne of CloseDistance;

S307, count the mean avgcatalog of the categories of all Cubes in the CubeList;

S308, obtain the distance index distanceindex, and its calculation formula is as follows:

S4. Calculate the priority of the homogeneous weights of each Cube in the CubeList:

S401, the following steps are performed for each Cube in the CubeList;

S402, grid currentCube to be processed at present=take out a Cube from the CubeList;

S403, the current center point centerpoint=coordinate cp of the center point position of the currentCube to be processed;

S404, neighbor list neighborlist=calculate the distance from the cp of all Cubes in the CubeList to the centerpoint, and extract all Cubes whose distance is less than or equal to distanceindex;

S405, allnum=number of cubes in neighborlist;

S406, the homogenization counter scounter=0, the homogenization counter ycounter=0, the iteration counter counter=1;

S407, testCube=take out the counter-th Cube in neighborlist;

S408, diff=ABS (the affiliation category of the currentCube—the affiliation category of the testCube);

Among them, ABS represents the absolute value of calculation;

S409, if the diff is greater than 2, then ycounter=ycounter+1, otherwise counter=scounter+1;

S410, counter=counter+1;

S411, if counter is greater than allnum, go to S412, otherwise go to S405;

S412, calculate the homogeneous connection permission pri, and the corresponding formula is as follows:

S413. The homogeneous weight priority=pri of the currentCube;

S5, continuously remove the three-dimensional space points from the PointDataSet according to the homogenous connection weights until the reduction percentage is reached:

S501, std=calculate the standard deviation of the priority of the homogeneous weights of all cubes in the CubeList;

S502, the rotation value circle=std/clusternum;

S503, sort all cubes in the CubeList from small to large according to the priority value;

S504, the smallest grid to be processed FirstCube=takes out the Cube ranked first in the CubeList;

S505, ptlist=in the PointDataSet, take out all the three-dimensional space points contained in the FirstCube space;

S506, if the number of points in the ptlist is less than 8, go to S507, otherwise go to S508;

S507, delete FirstCube from the CubeList, and go to S510;

S508, find the point in the ptlist closest to the coordinate cp of the center point of the FirstCube, and delete the point from the PointDataSet;

S509, the homogeneous connection weight priority of FirstCube=the homogeneous connection weight priority+circle of FirstCube;

S510, currentN=calculate the number of three-dimensional space points currently included in the PointDataSet;

S511, if (n-currentN)/n is less than percent, go to S503, otherwise go to S512;

S512, output the PointDataSet as the simplified result.

Example 2

A tree (Korean pine) consisting of points in 3D space:

Step 1, input a 3D pine tree composed of a point cloud, as shown in Figure 1; the PointDataSet in Figure 1 contains 1 million 3D points, the side length of the input grid is width=5, and the number of homogenized clusters is clusternum=30. Enter the simplified percentage percent=50%;

Step 2, the cube grid list CubeList;

Step 3, calculate and obtain the distance index distanceindex=0.7432;

Step 4: Calculate the priority of the homogeneous weights of each Cube in the CubeList;

Step 5, the simplified result is obtained, as shown in Figure 2; it can be seen that the number of three-dimensional space points is greatly reduced compared with the number in the original three-dimensional point cloud image, and the original shape of the tree is still guaranteed, indicating that this patent method is more effective.

Example 3

A tree (Smell pine) composed of three-dimensional space points using Embodiment 1 of the present invention

Step 1, input a three-dimensional pine tree composed of point clouds, as shown in Figure 3; the PointDataSet in Figure 3 contains 300,000 three-dimensional points, the side length of the input grid is width=4, and the number of homogenized clusters is clusternum=25. Enter the simplified percentage percent=50%;

Step 2, the cube grid list CubeList;

Step 3, calculate and obtain the distance index distanceindex=0.621;

Step 4: Calculate the priority of the homogeneous weights of each Cube in the CubeList;

Step 5, the simplified result is obtained, as shown in Figure 4, it can be seen that the simplification and the shape of the tree are still preserved.

Claims (1)

1. A tree-oriented three-dimensional point cloud data simplification method comprises the following steps:

s1, inputting a tree point cloud data set PointDataSet containing n three-dimensional space points, inputting a grid side length width, inputting a homogenization bunching number clusternum, and inputting a simplification percentage:

s101, inputting a tree point cloud data set PointDataSet, wherein the PointDataSet is a data set { p) formed by three-dimensional space points₁, p₂, …, p_nAny three-dimensional point pi = { px, py, pz }, px is the coordinate of the point on the x-axis, py is the coordinate of the point on the y-axis, and pz is the coordinate of the point on the z-axis;

s102, counting all points in the PointDataSet to obtain the maximum value xmax of an x coordinate, the minimum value xmin of the x coordinate, the maximum value ymax of a y coordinate, the minimum value ymin of the y coordinate, the maximum value zmax of a z coordinate and the minimum value zmin of the z coordinate of all points in the PointDataSet;

s103, x-axis span xdistance = xmax-xmin, y-axis span ydistance = ymax-ymin, and z-axis span zdistance = zmax-zmin of all points in the PointDataSet;

s104, inputting the grid side length width and the homogenization bunching number clusternum;

the default value of the grid side length width is xdistance/1000;

the default value of the homogenization bunching number clusternum is 100;

s2, cutting the three-dimensional space occupied by the PointDataSet into a plurality of cubic grids according to the side length width, and forming a cubic grid list cube:

s201, a cubic grid list CubeList = empty set;

s202, x iterator xcount = xmin;

s203, y iterator ycounter = ymin;

s204, z iterator zcounter = zmin;

s205, establishing a cubic grid Cube, wherein the Cube comprises the following attributes:

the spatial range scope corresponding to Cube is from (xcount, ycounter, zcounter) to (xcount + width, ycounter + width, zcounter + width);

coordinates of the position of the center point of Cube:

cp=(xcounter+width/2,ycounter+width/2,ycounter+width/2)；

cube point density = 0; cube's membership category = 0; cube's homogenous connection weight priority = 0;

s206, Cube point density = the number of all spatial points included in the Cube spatial range in the PointDataSet;

s207, adding the Cube into the Cube List;

S208，zcounter=zcounter+width；

s209, go to S210 if zcount > = zmax, otherwise go to S205;

S210，ycounter=ycounter+width；

s211, if ycounter > = ymax, go to 212, otherwise go to S204;

S212，xcounter=xcounter-1；

s213, go to S214 if xcount > = xmax, otherwise go to S203;

s214, the calculation process is ended;

s3, calculating the generic relationship category of each Cube in the Cube list to obtain the distance index:

s301, using the point density dense of each Cube in the Cube List as input data, dividing the input data into cluster categories through a k-means algorithm, and obtaining cluster centers c 1, c 2 and … c cluster;

wherein c 1 represents the clustering center with number 1, c 2 represents the clustering center with number 2, and c [ clusternum ] represents the clustering center with number clusternum; c [ i ] denotes the ith cluster center;

s302, for each Cube in the Cube List, calculating the distance between the density of the Cube and cluster centers, finding the cluster center closest to the density, and storing the serial number of the cluster center in the category relationship category of the Cube;

s303, establishing an array CloseDistance = null array;

s304, for each three-dimensional space point in the PointDataSet, finding another three-dimensional space point closest to the point and adding the distance value into the closeDistance;

s305, calculating the mean value avgdistacne of the CloseDistance;

s307, counting the average value avgcatalog of the categories of all the cubes of the Cube list;

s308, obtaining the distance index distanceindex, wherein the calculation formula is as follows:

s4, calculating the priority of each Cube homogeneous connection weight in the Cube List:

s401, processing the following steps for each Cube in the Cube list;

s402, taking out one Cube in the Cube List when the currentCube = the grid to be processed currently;

s403, the coordinate cp of the center point position of the center point center = current cube to be processed currently;

s404, calculating the distances from cp of all Cubes in the CubeList to centerpoint, and taking out all Cubes with the distances less than or equal to distanceindex;

s405, allnum = number of cubes in neighborist;

s406, the homogenization counter scount =0, the heterogeneous counter yzcount =0, and the iteration counter count = 1;

s407, testCube = take out the second counter Cube in neighborlist;

s408, diff = ABS (membership category of currentCube-membership category of testCube);

wherein ABS represents the calculated absolute value;

s409, yzcount = yzcount +1 if diff is greater than 2, otherwise scouter = scouter + 1;

S410，counter=counter+1；

s411, if the counter is larger than allnum, turning to S412, otherwise, turning to S405;

s412, calculating a homogeneous connection authority pri, wherein the corresponding formula is as follows:

s413. currentCube' S homogeneous weight priority = pri;

s5, continuously eliminating the three-dimensional space points from the PointDataSet according to the homogeneous connection weight value until the simplification percentage is reached:

s501, std = calculating the standard deviation of the homogeneous connection weight priority of all the cubes in the Cube List;

s502, a rotation value circle = std/clusternum;

s503, sorting all the cubes of the Cube List from small to large according to the priority values;

s504, taking out the first-ranked Cube in the Cube List, wherein the first-ranked Cube is the minimum grid Firstcube = to be processed;

s505, taking out all three-dimensional space points contained in the FirstCube space range within the pointDataSet by ptlist =;

s506, if the number of the ptlist midpoints is less than 8, turning to S507, otherwise, turning to S508;

s507, deleting the FirstCube from the CubeList, and turning to S510;

s508, finding a point nearest to a coordinate cp of the position of the center point of the FirstCube in the ptlist, and deleting the point from the PointDataSet;

s509, the homogeneous connection weight priority of the FirstCube = the homogeneous connection weight priority + circle of the FirstCube;

s510, currentN = calculating the number of the points of the PointDataSet which currently contain the three-dimensional space;

s511, if (n-current/n) is less than percent, go to S503, otherwise go to S512;

s512, outputting the PointDataSet as a simplification result.

Images